Advancements in Anomaly Detection and Deepfake Defense

The field of anomaly detection and deepfake defense is rapidly evolving, with a focus on developing innovative methods to detect and prevent manipulated multimedia content. Recent research has explored the use of wavelet transforms, multimodal models, and spatial-frequency aware fusion networks to improve detection accuracy and efficiency. These advancements have significant implications for industrial inspection, social media, and national security. Noteworthy papers include: Wavelet-Enhanced PaDiM for Industrial Anomaly Detection, which integrates wavelet analysis with convolutional neural networks to improve anomaly detection and localization. ERF-BA-TFD+, a multimodal model that combines audio and video features to detect deepfakes, achieving state-of-the-art results on the DDL-AV dataset. ClearMask, a noise-free defense mechanism that modifies audio mel-spectrograms to prevent voice deepfake attacks, demonstrating effectiveness against unseen voice synthesis models and black-box API services.

Sources

Wavelet-Enhanced PaDiM for Industrial Anomaly Detection

ERF-BA-TFD+: A Multimodal Model for Audio-Visual Deepfake Detection

Multi-scale Scanning Network for Machine Anomalous Sound Detection

Defending Deepfake via Texture Feature Perturbation

ClearMask: Noise-Free and Naturalness-Preserving Protection Against Voice Deepfake Attacks

FakeSV-VLM: Taming VLM for Detecting Fake Short-Video News via Progressive Mixture-Of-Experts Adapter

A Spatial-Frequency Aware Multi-Scale Fusion Network for Real-Time Deepfake Detection

Disruptive Attacks on Face Swapping via Low-Frequency Perceptual Perturbations

Built with on top of